Can routine chest CT add useful survival signals in pulmonary hypertension?
In 620 Sheffield referrals, automated right-atrial and coronary-calcification measurements were associated with all-cause mortality beyond established risk factors. The retrospective single-centre study does not show that using the measurements improves care or survival.
The short version
- What happened
- The retrospective Sheffield ASPIRE-registry analysis included 620 adults assessed between 2006 and 2020; 472 had pulmonary hypertension and 388 died during follow-up.
- Who it affects
- People with suspected pulmonary hypertension could eventually receive more information from a CT already obtained, without an extra scan, if the pipeline proves reliable across hospitals.
- What remains unproven
- Retrospective association study; no randomised or prospective decision-impact test
Keep in mind
Retrospective association study; no randomised or prospective decision-impact test
Single specialist centre, 80% white British and 78% GE scanners
Scans and referrals span 2006–2020, with varied acquisition and evolving treatment
All-cause mortality outcome does not establish pulmonary-hypertension-specific causation
Data-derived cut-offs need external validation and may be optimistic
No evidence that automated measurements improve treatment decisions, patient understanding or survival
Source analysis · 1 linked source. Primary analysed source date: 4 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 9 min The evidence, methods, implications and limitations
The direct answer: routine scans contained extra prognostic signals, not a proven treatment tool
A retrospective study found that automated measurements taken from ordinary non-gated, non-contrast chest CT scans were associated with survival among people referred for suspected pulmonary hypertension. Right-atrial volume and coronary-artery calcification remained linked to all-cause mortality after adjustment for age, sex and pulmonary-hypertension subgroup. In an additional analysis using the REVEAL 2.0 clinical risk score, the associations also remained statistically significant. The result suggests that scans already acquired during assessment contain quantifiable information that may complement established risk evaluation.
It does not show that an AI warning improves treatment, prevents death or should determine an individual prognosis. The analysis looked backward at a single referral-centre registry. Automated variables were related to deaths observed during follow-up, but clinicians were not randomised to use or ignore the measurements, and no change in care was tested. Coronary calcification and enlarged cardiac chambers are also established signs of cardiovascular burden and disease severity. Automation can make them easier to measure; it does not make their association causal.[1]
Who was studied and what counted as the outcome
Investigators identified 884 adults in the Assessing the Spectrum of Pulmonary Hypertension Identified at a Referral Centre, or ASPIRE, registry in Sheffield. Eligibility required suspected pulmonary hypertension, a right-heart catheterisation and a suitable non-contrast chest CT within one month. They excluded 264 people: 189 because the scan and catheterisation were more than a month apart, nine because survival status was missing and 66 because haemodynamic classification or disease subtype was unavailable. The final denominator was 620 people enrolled from 2006 to 2020.
The cohort's mean age was 64 years with a standard deviation of 13, and 62% were female. Four hundred and seventy-two people had pulmonary hypertension: 202 pulmonary arterial hypertension, 84 disease associated with left-heart disease, 112 associated with lung disease, 65 chronic thromboembolic disease and nine unclear classification. Another 148 did not have pulmonary hypertension after assessment. During follow-up, 388 people died, or 63% of the cohort. Survival ranged from two to 5,187 days, with a mean of 1,659 days. The endpoint was all-cause mortality, not a pulmonary-hypertension-specific death.[1]
The distinction that matters
What the retrospective pipeline did—and did not test
The study turned existing CT scans into measurements, then related those measurements to later all-cause mortality.
Cohort
620 Sheffield referrals with CT and catheterisation
Single-centre retrospective sample
Automation
Segment chambers, vessels and coronary calcium
Quantitative imaging variables
Analysis
Adjust associations for demographics, disease group and REVEAL
Independent statistical signals
Unproven step
Use measurements to change care
No tested improvement in decisions or survival
Two validated models measured structures and calcium on non-dedicated CT
The researchers applied two previously validated deep-learning systems. One segmented cardiothoracic structures including the combined ventricles, right and left atria, pulmonary artery, ascending aorta and descending aorta. The other measured calcification in the right coronary artery, the main stem and left anterior descending artery together, and the left circumflex artery. These were non-gated scans taken without contrast, so they were not dedicated coronary-calcium or cardiac examinations. The practical proposition is that a pipeline could recover structured cardiovascular measurements without ordering a new scan.
Calcification was present in 45% of participants in the right coronary artery, 67% in the main-stem and left-anterior-descending territory and 35% in the left circumflex artery. People with pulmonary hypertension had larger average right-atrial, ventricular and pulmonary-artery volumes than the group without the disease. Those group differences describe this referral cohort and do not establish a diagnostic threshold. For the mortality models, structural volumes were analysed per 10-millilitre increase and calcification presence was treated as a predictor, with separate adjusted models rather than a single deployed clinical score.[1]
The strongest adjusted associations involved coronary calcification
After adjustment for age, sex and pulmonary-hypertension subgroup, each 10-millilitre increase in right-atrial volume was associated with a hazard ratio of 1.024, with a 95% confidence interval from 1.009 to 1.039. In plain language, the model estimated about a 2.4% higher instantaneous risk of death per 10 millilitres, conditional on the included variables. Calcification was associated with hazard ratios of 1.378 for the right coronary artery, 1.868 for the main-stem and left-anterior-descending territory and 1.342 for the left circumflex artery. Their confidence intervals excluded one.
The paper also converted measurements into high and low groups using cut-offs selected from the same data with the Youden index. A right-atrial volume above 117.95 millilitres had an adjusted hazard ratio of 1.611. Calcification above 0.01 millilitres had hazard ratios of 1.533, 2.114 and 1.465 across the three coronary territories. These thresholds may make graphs easy to read, but data-derived cut-offs can look better in the development cohort than in new patients. They require external validation before they can be treated as clinical decision boundaries.[1]
Adjustment for REVEAL does not prove added clinical usefulness
The researchers asked whether the imaging variables remained associated with mortality after adjustment for REVEAL 2.0, an established clinical risk score used primarily in pulmonary arterial hypertension. Right-atrial volume remained significant with a reported hazard ratio of 1.026 per 10 millilitres. Coronary-calcification associations were 1.604 for the right coronary artery, 2.339 for the main-stem and left-anterior-descending territory and 1.576 for the left circumflex artery. A pulmonary-artery-to-aorta ratio was no longer significant in that analysis, suggesting overlap with information already captured by clinical risk assessment.
A variable remaining statistically significant after adjustment is not the same as demonstrating better decisions. The paper did not report a prospective comparison of discrimination, calibration, reclassification benefit and clinical utility under a locked workflow. Nor did it test whether clinicians would act differently or whether those actions help patients. A model can add a statistically independent association while providing little useful separation between individuals, or while prompting interventions that do not improve outcomes. Prospective decision-impact evidence is therefore the key missing step.[1]
Single-centre data and older scans constrain generalisation
All patients came from one specialist centre in Sheffield, and 80% were recorded as white British. Seventy-eight per cent of scans came from GE systems, 18% from Canon and 4% from Siemens. Although every scan met predefined quality criteria, the examinations spanned 2006 to 2020 and used varied acquisition and reconstruction settings. The model therefore encountered technical variation, but not the institutional, demographic and workflow diversity of a genuinely external multi-centre validation. Contemporary treatment patterns may also differ from those experienced by the earliest participants.
The retrospective design leaves room for residual confounding. Age, sex and disease subgroup are important, and REVEAL captures additional clinical information, but unmeasured comorbidities, treatment changes and referral patterns can influence survival and imaging markers. The study did not use an a priori power calculation, and testing several anatomical variables and derived cut-offs increases the risk of optimistic findings. The high death count supplies substantial statistical information, yet it cannot remove bias created by selection or measurement choices.[1]
How the finding could affect patients and imaging teams
If the measurements are independently validated, a hospital might use them to enrich a pulmonary-hypertension assessment without exposing a person to more radiation or requiring a dedicated cardiac scan. A radiologist could receive consistent chamber and calcium measurements, while the pulmonary-vascular team could decide whether an unexpected signal warrants closer review alongside symptoms, exercise capacity, biomarkers, catheterisation and other imaging. The potential human benefit is earlier recognition of a person whose overall cardiovascular burden may be greater than a visual report conveys.
The risks are equally practical. A mortality-associated marker can be frightening if presented as a personal prediction without calibration or context. Automated segmentation may fail silently on unfamiliar anatomy, motion, scanner settings or severe disease. A high-risk flag could cause extra appointments or tests without changing useful treatment, while a reassuring output could distract from clinical deterioration. Any deployment should show the source images, measurement quality and uncertainty, allow correction, and make clear that the tool supplements rather than replaces specialist assessment.[1]
What would change the assessment
The next step should be external validation in hospitals with different referral thresholds, ethnic composition, scanner fleets and current treatment pathways. Researchers should lock the models and thresholds before testing, report calibration and missing or failed segmentations, compare performance against the full clinical workflow, and evaluate groups separately without underpowered claims. A prospective impact study should randomise or otherwise robustly compare teams with and without the automated measurements, then measure changes in decisions, unnecessary testing, treatment timing, patient understanding and outcomes.
Data on mortality are consequential, but urgency depends on an active event or a time-sensitive protective action. This paper describes associations in historical scans and does not announce a care change, product recall, safety warning or immediate intervention. It belongs in the current Research & Impact feed, not the Breaking desk. The assessment would strengthen only if external prospective evidence shows that the automated information is reliable, improves decisions beyond established tools and ultimately benefits people rather than merely producing additional risk numbers.[1]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring clinical ai: compare the finding and its evidence limits.
Can an MRI model predict cancer progression? A 33-patient holdout is promising, not treatment guidance
A Chinese single-centre model separated progression risk after chemoradiotherapy in 113 patients, but the holdout contained only 33 people and no external hospital. The paper's feature count is also internally inconsistent.
Read the next reportExplore the full source and evidence record
What this means for people
- People with suspected pulmonary hypertension could eventually receive more information from a CT already obtained, without an extra scan, if the pipeline proves reliable across hospitals.
- Clinicians need measurement quality, uncertainty and source images—not an unexplained mortality label—to judge whether an automated result is credible.
- The study did not show that displaying these markers changes care, reduces unnecessary testing or improves survival.
Global context
Pulmonary-hypertension services and CT protocols vary widely. Reusing existing scans could help resource-constrained systems, but a Sheffield referral cohort cannot establish accuracy or clinical benefit across regions without diverse external validation.
What the evidence does not yet show
- Retrospective association study; no randomised or prospective decision-impact test
- Single specialist centre, 80% white British and 78% GE scanners
- Scans and referrals span 2006–2020, with varied acquisition and evolving treatment
- All-cause mortality outcome does not establish pulmonary-hypertension-specific causation
- Data-derived cut-offs need external validation and may be optimistic
- No evidence that automated measurements improve treatment decisions, patient understanding or survival
What to watch next
- Locked-model external validation across centres, scanner vendors and patient groups
- Calibration, segmentation failures and performance at clinically relevant thresholds
- Incremental utility beyond the complete REVEAL workflow, not significance alone
- Prospective effects on decisions, testing, treatment timing and patient outcomes
- Clear communication that a risk marker is not an individual certainty
Living evidence record
Impact record IAI-1JHT47R
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
4 November 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what European Radiology Cardiovascular and Thoracic published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 4 November 2026
Research question, topics and regions
Whether automated measurements of cardiothoracic structures and coronary calcification on routine non-contrast chest CT are associated with mortality in people referred for suspected pulmonary hypertension
Topics: Pulmonary hypertension · Chest CT · Risk stratification · Cardiovascular imaging · Clinical AI
Regions: United Kingdom · Europe · Global
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.